How does Siri know what you mean? You could say “What do I have on my calendar today?” or “How does my day look today?” or “Am I busy today?”, and countless additional combinations of words that could mean the same thing to any English speaker, but how does Siri, a piece of a computer program, know what you mean? After all, it’s practically impossible to enter beforehand every possible question about your calendar and schedule into Siri. Actually, that would be literally impossible. Not only that, how does Siri know what you’ve said? From a random audio file sent from your phone to Apple’s servers, the program is able to very accurately determine what you meant, and this can be done for millions of variations in voice pitch, tone, accents, and length. This is just one example of artificial intelligence that is slowly populating our lives.
Traditional computer programs act in a consistently predictable way, taking action X when given a certain input, action Y when given another input, and Z when given yet another input. That’s enable computers and robots to replace many repetitive jobs that people have had to do since the beginning of time. But what about things like organizing music libraries into tastes, understanding speech, driving, and vision? Those more complex actions can’t be programmed the traditional way, because there are an infinite number of possible inputs, and equally numerous possible actions that would need to be taken. For a long time, these tasks were thought to be uniquely human. But we are, after all, nothing but a series of electrical signals moving in parallel, so theoretically, replicating our entire perceptive system digitally should be possible*. We haven’t done that exactly, yet. But we’re getting closer, and developments in that area has been unlocking potentials for machines that we could not have thought was possible just several decades ago. In simple terms, A.I., or artificial intelligence, is a general term referring to any replication of intelligence and learning in a computer, but a model called Deep Learning is the most common today.
Now, computers are, by many definitions, intelligent.
Deep Learning works by simulating on a computer a radically simplified model of the human brain. In essence, the computer simulates a connected network of “neurons” that can be switched on or off according to the input, and the relationships between the neurons’ connections and on/off state produce a resulting set of neurons that are switched a certain way, which in turn corresponds to an output from the program. This allows the program’s input and output to be quite a bit more “flexible”, and it also allows the program to “learn” on its own. What this means is that, instead of a programmer putting into the computer which output should correspond to every possible input, the program can, on its own, look at a set of data and determine those connections. It’s for this reason that computers can now recognize certain objects and understand spoken words, as well as the meanings behind them. In short, now, computers are, by many definitions, intelligent.
Artificial intelligence is not a remote concept at all to most of us. Even if you don’t talk daily into your smartphone’s personal assistant, things like YouTube’s video suggestions, your e-mail’s spam filters, Amazon’s tailored ads, and song recognition by services like Shazam are powered by the same Deep Learning computer algorithms. The computer are learning to distinguish the videos you like from those you don’t, learning to distinguish spam from relevant mail, and learning which items you are most likely to purchase. But the potential extends well beyond just simple tailoring of ads or suggestions. A.I. vision algorithms allow people to create software that detects signs of cancer that physicians miss, cars that can tell pedestrians apart from trees or road signs, and even smartphone apps that recognize algebra problems and shows you how to do it from just a picture of the problem. And those are just the things machine learning can do right now, from devices that fit in our pockets. As with all aspects of technology, the things that machine learning will allow us to do in the future is only limited by our imagination. With computers that can churn along 24/7 for months, it’s not far-fetched at all to envision a computer system that supersedes our abilities of perception and reasoning. But when that day comes, what happens to intelligence? Does it lose its meaning as an inherently human quality?
Naturally, nobody will be able to provide a satisfactory answer until we see computers that really are indistinguishable from humans. But here’s my take: yes, intelligence won’t be inherently human, and computers, for all intents and purposes, will be indistinguishable from people in the ability to reason and even feel emotions. That day may not come for quite a while, but it’ll come eventually. And when it does, the traditional notion of what defines us will have to change. But I don’t think that emergence of more human computers will somehow de-value the uniqueness of people like many suggest. On the contrary, I think the fact that we can create computers that can think about the world on our level of complexity is a mirror of our deep understanding of who we are, and regardless of how you look at it, that’s a step in the right direction. If anything, developments will help us understand what intelligence is, and who we are as people of intelligence. But I think intelligence is hardly the trait that defines us as people; rather, it’s the fact that we can communicate, collaborate, and create together something on the scale of our own minds that define us. The boundaries of intelligence may be getting blurry, but ultimately, I hope that’s not where our identities lie.
* Possible, but excruciatingly slow. It takes hundreds of hours of time at a computer to simulate a fraction of a second’s actions in a structure as complex as the human brain.
← The hashtag poets
Thoughts for 2014 →
I share new posts on my newsletter. If you liked this one, you should consider joining the list.
Have a comment or response? You can email me.